banana-claude

Banana Claude turns a plain-language creative request into a planned and reviewable Gemini image workflow. It shows the prompt, model, data settings, output destination, and nominal estimate before any paid request.

Here's the skill, here's what it does, here's how to install it.

Start here

Requirements: Claude Code 2.1.199 or newer, Python 3.11 or newer with python3 on PATH, and a Gemini API key for a billing-enabled Google AI project.

/plugin marketplace add AgriciDaniel/banana-claude
/plugin install banana-claude@banana-claude-marketplace
/plugin enable banana-claude@banana-claude-marketplace
/reload-plugins

Then ask for the image you want:

/banana-claude:banana generate an urban 16:9 GitHub hero with clean left-side copy space

Banana Claude plans offline first. Review the exact plan, then approve the paid request when it matches your intent.

The workflow is simple

1. Ask: describe the outcome in normal language.
2. Plan: Banana Claude builds the visual brief and chooses a route.
3. Review: inspect the prompt, references, privacy settings, and estimate.
4. Approve: one short-lived approval authorizes one provider attempt.
5. Create: the image and a privacy-conscious metadata sidecar are saved.
6. Check: review the actual pixels, then fix, regenerate, or ship.

What you can make

• Generate: campaign visuals, covers, product scenes, diagrams, concepts, social assets, and more.
• Edit: make one clear change while protecting identity, geometry, and brand details.
• Continue: iterate with stored Flash or Pro sessions, or attach the last result as a fresh reference.
• Compare: test up to three prompts across three model routes in one bounded portfolio.
• Review: check copy, crop, composition, consistency, artifacts, rights, and provenance.
• Typeset: add approved copy, fonts, logos, and trusted raster art locally when exact text matters.

Banana Claude routes across Google's current Nano Banana 2 Lite, Nano Banana 2, and Nano Banana Pro models. See the dated Gemini model reference for current capabilities, pricing, and the deprecated compatibility route.

Built for control

• The plugin installs disabled because image generation is a paid service.
• API keys stay out of commands, request URLs, metadata, and public CI.
• Every paid tool shows its plan and requires a user decision before execution.
• One approval permits one provider attempt. A changed plan needs a new approval.
• Cost records stay private and omit raw prompts by default.
• Generated output is never treated as automatically correct or production ready.

Gemini output is probabilistic. Banana Claude cannot guarantee consistency, spelling, geometry, policy acceptance, rights clearance, or a final invoice. Those remain review conditions, not marketing claims.

Read Security and privacy before production use.

Upgrade from public 1.4.1 or 2.1.0 installs

Do not simply overwrite an older public install. It may have left an unpinned third-party MCP entry, a raw Google key in Claude settings, or an obsolete skill directory. Version 3 includes a redacted scan, review-first cleanup, and explicit state migrations.

Follow the upgrade guide before enabling version 3. Rotate any key that an older setup stored on disk or exposed in shell history.

Documentation

• User guide: installation, upgrades, workflows, and direct scripts
• Prompt craft: visual briefs, consistency, composition, and recovery
• MCP tools: planning and execution contracts
• Security: credentials, approvals, outputs, state, and reporting
• Changelog and version 3 release notes

Contributing

Contributions are welcome. Read CONTRIBUTING.md before opening a pull request. For help or a reproducible bug, open an issue without keys, private media, or full environment dumps.

License

MIT License, see LICENSE.

Built by Agrici Daniel, AI Workflow Architect.

Blog · AI Marketing Hub · YouTube · More open-source projects